Four Years of Generative AI
Fascination, denial, delusion, panic. Usama Fayyad traces the stages organizations have moved through since generative AI arrived.
AI in Action • On-Demand Library
Watch conversations from AI in Action and explore resources from Northeastern University experts and business leaders working through the opportunities and questions shaping AI today.
Find recordings, expert Q&As, presentations, and materials from the year-round AI in Action series and Business Summit.
New content is added as the series continues.
AI in Action • October 2026
Full conversation
After four years of generative AI, many organizations are still deciding which projects deserve investment and how fast they should deliver. In this AI in Action conversation, Usama Fayyad and Tom Davenport discuss how the market has reacted to generative AI, how leading companies decide which AI projects should exist at all, and why timelines matter.
Featuring Usama Fayyad and Tom Davenport.
Short clips
The Generative AI Cycle
Fascination, denial, delusion, panic. Usama Fayyad traces the stages organizations have moved through since generative AI arrived.
Prioritizing AI Investments
Tom Davenport explains why Capital One asks whether an AI project has a right to exist before it ever gets to return on investment.
Scoping AI Projects
Usama Fayyad lays out his 12-week rule for AI projects and what it means for how organizations scope and commit to new initiatives.
From the AI Ready or Risky? working session
Executives at this event pressure-tested a real AI initiative across four areas: business value and ROI, people readiness, data readiness, and governance and access. Use the same card to see where your initiative is ready, where risk remains, and what needs to happen next.
AI in Action • September 2026
Full conversation
Giving employees access to AI tools does not automatically change how they work. In this AI in Action conversation, Zach Blattner and Berkeley Almand-Hunter discuss what helps AI learning stick, including hands-on practice, role relevance, time to experiment, and how organizations can think more clearly about real adoption.
Featuring Zach Blattner and Berkeley Almand-Hunter.
Short clips
Employee AI Adoption
Access alone does not create adoption. This clip looks at why organizations need more than licenses and logins if they want employees to actually use AI well.
Learning and Practice
Employees often want AI to help them, but they also need room to practice, test ideas, and build confidence before new habits take hold.
What Impact Looks Like
What should organizations actually measure? This clip explores why true adoption goes beyond simple usage counts and should connect back to work outcomes.
Featured conversation
AI in Action • July 29, 2026
Why 2026 became the year of AI panic, what it means for entry-level hiring, how fast agents are spreading, and why your own data decides what AI can do for you.
Featuring Usama Fayyad and Liz Kohler.
From AI Panic to AI Progress • Audience Q&A
Questions poured in before and during this conversation, and many asked the same thing in different words. We grouped them into the 12 answers below from Usama Fayyad. Open any answer to see the questions it covers.
32 people asked about this
Don’t boil the ocean. Start with a narrow slice of a high-friction, high-effort process that can move a meaningful business metric. Deliver a clear result in roughly 12 weeks, use only the data that slice needs, and build on an architecture that can support the next slice. Short cycles create visible value and pull from the business. Organizations that get past experimentation also invest in workforce education, culture, trusted expertise, and responsible-AI risk.
From Usama’s follow-up Q&A session
17 people asked about this
Treat ROI as multidimensional, not just labor reduction. Measure time and effort saved, speed, quality, consistency, customer experience, and risk and compliance exposure. Set an honest baseline, and bring in finance, legal, risk, operations, and business owners, because a use case that looks efficient to one group can create cost or risk for another.
From Usama’s follow-up Q&A session
11 people asked about this
Model pricing is opaque, and token consumption can blow up a budget. Avoid depending on one vendor, route work to the least expensive model that meets the need, and escalate to larger models only when quality requires it. Open-weight and smaller models, including privately hosted ones, can cut costs dramatically.
From Usama’s follow-up Q&A session
32 people asked about this
Fear is a culture and change-management issue, and the answer is education, especially experiential learning tied to people’s actual work. Let employees compare today’s process with an AI-enabled version, practice on real projects, and see how their role becomes more valuable. That makes learning relevant and turns a perceived threat into a capability.
From Usama’s follow-up Q&A session
24 people asked about this
AI will eliminate some tasks and jobs, but entry-level work isn’t disappearing. The bigger issue is a mismatch between traditional training and what employers now need. New graduates and junior staff who can apply AI pragmatically will be highly valuable, and curricula should evolve to teach practical use of current tools.
Watch Usama answer at 15:20 ↗16 people asked about this
Don’t mistake fluency for understanding. Models make errors, are sensitive to their data, and can give one confident answer that hides the alternatives. Use them where some error is tolerable, require stronger human review in regulated or high-stakes work, protect sensitive inputs, and learn on lower-risk uses where the technology’s strengths and limits become clear.
From Usama’s follow-up Q&A session
13 people asked about this
Use narrow, focused models when they meet the requirement. They’re cheaper, easier to fine-tune, can often run privately, and expose less data and compute. Build infrastructure that can swap among models instead of locking into one vendor, which improves sovereignty, negotiating power, privacy, and cost control. Expect agentic AI and smaller, routable models to become standard.
Watch Usama answer at 34:08 ↗6 people asked about this
Data is a first-order blocker. General-purpose models don’t know an organization’s proprietary processes or context, and most enterprise data was prepared for human summaries, not the granular form algorithms need. Make use-case-specific data available and reliable instead of assuming the model already knows the business. Lasting advantage comes from that proprietary data and process knowledge, not from access to a frontier model.
Watch Usama answer at 2:47 ↗8 people asked about this
Don’t assume frontier models or enterprise subscriptions make proprietary inputs risk-free. Sanitize, anonymize, and minimize what you share. For sensitive documents and retrieval (RAG) systems, keep data and models in an environment your organization controls when possible, which preserves legal, policy, and technical safeguards and limits exposure of customer data, IP, and trade secrets.
Watch Usama answer at 32:32 ↗3 people asked about this
AI safety spans responsible AI, privacy and civil liberties, cybersecurity, IP leakage, and resistance to attacks that try to extract protected information. Usama expects a major AI-related incident, most likely involving cybersecurity or a significant data leak, and notes that vulnerable legacy infrastructure raises the risk.
Watch Usama answer at 9:43 ↗11 people asked about this
Agentic AI is reaching broad scale and will touch virtually every industry. Agents won’t only automate whole roles; they’ll help with scheduling, service, research, troubleshooting, and coordinating workflows. Even hands-on occupations will use agents to access knowledge and improve how work gets done.
Watch Usama answer at 23:19 ↗2 people asked about this
AI already touches every stage of research: literature review, finding gaps, developing hypotheses, testing, writing, and review. Work that took weeks or months can sometimes take hours. But automated paper generation and review can degrade scientific quality, so researchers need to accelerate discovery without giving up judgment or the training of the next generation.
Watch Usama answer at 27:17 ↗Answers summarize Usama Fayyad’s remarks in the live session and a follow-up Q&A recording; they are not word-for-word quotes. Questions are lightly edited for clarity. “Related” questions are ones his answer speaks to without fully resolving.
AI in Action 2026 Business Summit
Ahead of the summit, Northeastern University alumni shared the questions they were asking about AI. We brought those questions to business leaders, researchers, entrepreneurs, and AI experts.
Watch their responses below.
Career & Workforce
AI in Action experts discuss how AI may change entry-level work, why early adoption matters, and how people can increase their value by learning to work effectively with AI.
Risk & Opportunity
AI in Action experts move beyond the extremes to discuss AI's opportunities, risks, regulation, and why fear should lead to better choices rather than paralysis.
AI Adoption
AI in Action experts discuss transparency, culture change, everyday workflows, and human oversight in moving AI from occasional use to meaningful adoption at work.
Public Impact
Hear examples of AI supporting under-resourced public services, including legal work and efforts to identify people who may be missing benefits or services they are entitled to receive.
Data Strategy
AI in Action experts explain why proprietary company data can be a major competitive asset, especially when it is fragmented, inaccessible to large models, or underused internally.
AI Skills
Experts distinguish between building AI and becoming a strong user of it, emphasizing AI literacy, fluency, experimentation, and the value of learning across disciplines.
Education
AI in Action experts discuss how AI could reshape higher education, why learning still matters, and how colleges may need to rethink what and how they teach.
State of AI
Experts cut through the hype to discuss AI's current reality, where its impact may be underestimated, and what organizations and individuals should be doing now.
Getting Started
AI in Action experts share advice on experimenting with AI, building daily habits, learning across disciplines, and becoming more effective in an AI-enabled workforce.
From the 2026 Business Summit
Explore Usama Fayyad's presentation on AI productivity, emerging developments, and considerations for organizations deciding where and how to apply AI.
AI in Action • 2026 Outlook
Before 2026 began, Usama Fayyad wrote down 13 predictions for the year: AI-driven work acceleration at scale, a reckoning with the technical debt piling up from prompts and vibe coding, smaller and cheaper models, data as the hardest blocker, and agentic workflow automation becoming real.
He originally shared them with Northeastern University leadership. Read the full list and score how they are holding up.
Work with Northeastern University
Beyond these conversations, Northeastern University works with organizations on workforce learning, finding where AI can create value, building and testing solutions, and using AI responsibly. Start with a free resource, or talk with the team that fits your goal.
01
Build workforce capability
Build practical AI skills around the tools, workflows, and challenges your employees encounter every day.
Explore Custom Learning ↗02
Identify AI opportunities
Examine business challenges, workflows, data, feasibility, potential value, and risk before committing to a larger build.
Explore free resources
AI Readiness Card → IdeaJAM: Design Thinking & AI Integration → Usama Fayyad’s AI predictions for 2026 →03
Build and test
Scope, prototype, test, and develop AI applications around your organization's workflows, users, data, and technical environment.
Explore the AI Solutions Hub ↗ Or explore Working Labs ↗04
Use AI responsibly
Build governance, risk review, and responsible AI practices that keep pace as AI use grows across your organization.
Free help
Ask an AI expert a question →Continue the conversation
Tell us what you’re working on, and we’ll connect you with the Northeastern University team best suited to help.